speedystats

speedystats is a Python package designed to accelerate NumPy statistical operations using Numba. The package maintains a clean API that mirrors NumPy's interface while providing significant performance improvements through parallel processing.

"Is it possible to use numba to speed up my median computation?"

Yep. Just use speedystats.

"Is it possible to use numba to make np.std go faster?

Yep. Just use speedystats.

Features

• Drop-in replacement for common NumPy statistical functions
• Significant performance improvements through Numba optimization
• Maintains NumPy-like API for easy integration

Installation

pip install speedystats

Quick Start

import numpy as np
import speedystats as fs

# Create some test data
data = np.random.randn(1000, 1000)

# Use just like numpy
mean = fs.mean(data, axis=0)
std = fs.std(data, axis=1)
median = fs.median(data)

# Works with nan values too
data_with_nans = data.copy()
data_with_nans[0, 0] = np.nan
nanmean = fs.nanmean(data_with_nans, axis=0)

Available Functions

• Basic Statistics: mean, median, std, var, sum
• Range Statistics: ptp (peak-to-peak)
• Percentile Functions: percentile, quantile
• NaN-aware Variants: nanmean, nanmedian, nanstd, nanvar, nansum
• Additional Functions: average, zscore

Performance Note

While speedystats is designed for performance, the actual speedup depends on your specific use case, data size, and hardware. The package is most effective with:

• Large arrays (typically > 100,000 elements)
• Multi-core processors (for parallel execution)
• Certain methods are sped up much more than numpy
• Certain axis / dimension combinations get huge speedups, others are usually comparable to numpy

Note: Benchmarking tools for automatic routing to speedystats implementation vs numpy default methods isn't finished --- so you are responsible for determining whether speedystats is faster. Here's an example of how to test it quickly:

from time import time
import numpy as mp
import speedystats as fs

# Suppose you want to test if median is faster for arrays of a certain shape and size
array = np.random.randn(1000, 10000)
axis = 1

# Set repeats to get a better estimate
num_repeats = 20

# numba functions always have to be compiled, which takes a moment,
# but speedystats caches them so you only have to wait once (it's just a second or so)
# so it's good to use a "warmup":
# you'll need to warmup each combination of number dimensions (2 here) and axis
_ = fs.median(np.zeros((10, 10)), axis=axis)

t = time()
for _ in range(num_repeats):
   _ = np.median(array, axis=axis)
numpy_time = time() - t

t = time()
for _ in range(num_repeats):
   _ = fs.median(array, axis=axis)
speedystat_time = time() - t

print("Speedup: ", numpy_time / speedystat_time)

Comprehensive benchmarking tools are under development. When they're released, this will be automated.

Development Status

This package is in beta. While the core functionality is stable, we're actively working on:

• Comprehensive benchmarking suite
• Performance optimization guides
• Additional statistical functions
• Advanced documentation

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the GNU General Public License v3 (GPLv3) - see the LICENSE file for details.